News 8 May, 15 May, 22 May 2023 - simon-oz/Weekly-AI-news GitHub Wiki

Welcome to the Weekly-AI-news wiki! I mainly focuses more on Generative AI, LLMs and the related, so may miss import AI news in other news.

I added the following previous weeks' news since May 2023

8 May 2023

  1. A.I. Is Getting Better at Mind-Reading - In a recent experiment, researchers used large language models to translate brain activity into words. Accuracy can now reach about 83% based the authors’ experiments.

  2. Thursday, Microsoft announced AI powered Bing plugins, enable Bing to search/generate multimedia content, and an Action feature coming soon.

  3. Google: "We have no moat, and neither does OpenAI". It’s reported that Google is considering to follow the trend of open source of LLM, such as LLAMA. Many startups and company released their AI Chat bot simply based on LLAMA.

15 May 2023

  1. Google.io conference – 1) Introduced PaLM v2 which support Google’s Bard, improved programming coding and dialog quality of Bard, support over 100 language translation, image <-> text generation/analysis, and more; 2) Google search supercharged with AGI, multi-round QA allows following-up question; 3) improved Gmail and Google photos features and others
  2. Anthropic’s Claude model can process 100K tokens now, more than 3X bigger than GPT-4’s 32K tokens.
  3. Microsoft announced new capabilities in Copilot, including semantic index, copilot in whiteboard makes Teams meetings and brainstorms more creative and effective; integrate DALL.E into PowerPoint to automatically generate ppt slides.
  4. OpenAI announced Shape-E, a conditional generative model for 3D assets. Type in text prompts into Shape-E, and the model will produce 3D objects that create better more detailed, and accurate objects.
  5. HuggingFace announced Transformers Agent – the agent provides a natural language API, which can interpret natural language and use a list of curated tools. The agent dramatically simplifies the process of a pre-trained LLM to call tools. A similar function with LangChain – a framework for developing applications powered by LLM.
  6. Meta announced ImageBind – an AI model capable of binding data from six modalities at once, including the 3D shape of an image. ImageBind outperforms prior individually trained models and helps AI by enabling machines to better analyze many different forms of information together.
  7. OpenAI applied GPT-4 to automatically propose explanations for GPT-2’S 300K neurons, and found neurons responding to concepts like similes, “things done correctly”, or expressions of certainty. GitHub link.
  8. IBM announced Dromedary, another LLM which uses principle-driven, self-alignment to minimize human supervision, and surpasses the performance of ChatGPT and Alpaca.

22 May 2023

  1. On 13 May, Sam Altman announced on his Twitter that “all ChatGPT Plus users getting browsing and plugins over the next week”. The time AI can use tools is coming. A new web-browsing feature is set to allow ChatGPT-Plus users to access real-time information. They will also get access to more than 70 plug-ins on sites including Expedia and Instacart (https://www.businessinsider.com/chatgpt-openai-web-browsing-plug-change-how-we-use-internet-2023-5).
  2. LangGPT — Empowering everyone to become a prompt expert! LangGPT addresses how to write high-quality prompts, which is becoming more akin to programming in the AI ear. The project link is here.
  3. On 17th May, Google announced AI-powered features will add to Colab, and free of charge. The features include code completions, natural language to code generation and even a code-assisting chatbot.
  4. On 18th May, OpenAI announced ChatGPT app for iOS. The ChatGPT app is free to use and syncs users history across devices. It also integrates Whisper, an open-source speech-recognition system, enabling voice input. ChatGPT Plus subscribers get exclusive access to GPT-4’s capabilities, early access to features and faster response times.
  5. OpenAI CEO calls on government to regulate AI - OpenAI CEO Sam Altman testified before the Senate Judiciary Committee on Tuesday, calling on Congress to pass legislation to regulate AI.
  6. Drag your GAN – a technology allow one to "drag" any points of the image to precisely reach target points in a user-interactive manner.